CoolFace
Modelpublic

Beidouqixing/minimax-h3-4step-lora-flashgen

sourceHugging Faceapache-2.0updated 2d agoView on Hugging Face
21likes4.7kdownloads
Model Card

MiniMax-H3 4-Step LoRA (FlashGen)

4-step distilled LoRA for MiniMax-H3 text-to-video+audio (T2VA). Trained on Ascend NPU with data-free distillation; merge into the base model and inject base_schedule before vllm-omni / MindIE-SD deployment.

Files

FileDescription
minimax_h3_4step_lora_flashgen_v1.0_768p_bf16.safetensorsStudent LoRA, rank 64, bf16
merge_lora_ckpt.pyMerge LoRA into base model and inject base_schedule

Training config

ItemValue
MethodDistribution matching distillation (VSD, no GAN), trained on Ascend NPU
TaskT2VA
Inference steps4
Resolution1344×768, 5.2s
base_schedule[1.0, 0.7, 0.4, 0.15, 0.0]
LoRA targetsqkv_proj, out_proj, fc1, fc2, adaln_proj.linear
LoRA rank64

Usage

Requires base model MiniMaxAI/MiniMax-H3.

Merge with merge_lora_ckpt.py before vllm-omni / MindIE deployment:

bash
python merge_lora_ckpt.py \
  ${MODEL_ROOT}/FL2VA \
  minimax_h3_4step_lora_flashgen_v1.0_768p_bf16.safetensors \
  ./minimax-h3-4step-flashgen

export MODEL=./minimax-h3-4step-flashgen/FL2VA

Deployment (MindIE-SD / vllm-omni)

For Ascend NPU inference and serving, see the MindIE-SD deployment guide: